Named Entity Recognition

NER.r

options(java.parameters = "- Xmx1024m")
#load libraries
library(openxlsx)
library(rJava)
library(NLP)
library(openNLP)
library(RWeka)

#read text
text <- c("My name is Catherine Gitau, I work at Ongair Limited in Nairobi, Kenya")

#convert the character vectors into one character vector
text <- paste(text, collapse = " ")
print(text)

#converts bio variable into a string
text<- as.String(text)

#create annotators for words and sentences
word_ann <- Maxent_Word_Token_Annotator()
sent_ann <- Maxent_Sent_Token_Annotator()

#Identifies where the sentences are and the words
text_annotations <- annotate(text, list(sent_ann, word_ann))
head(text_annotations)

#combines bio and the annotations
text_doc <- AnnotatedPlainTextDocument(text, text_annotations)
words(text_doc)

#creates annotators of kind person, location and organization
person_ann <- Maxent_Entity_Annotator(kind = "person")
location_ann <- Maxent_Entity_Annotator(kind = "location")
organization_ann <- Maxent_Entity_Annotator(kind = "organization")

#holds annotators in the order to be applied

pipeline <- list(sent_ann,
                 word_ann,
                 person_ann,
                 location_ann,
                 organization_ann)
text_annotations <- annotate(text, pipeline)
text_doc <- AnnotatedPlainTextDocument(text, text_annotations)

# Extract entities from an AnnotatedPlainTextDocument
entities <- function(text, kind) {
  s <- text$content
  a <- annotations(text)[[1]]
  if(hasArg(kind)) {
    k <- sapply(a$features, `[[`, "kind")
    s[a[k == kind]]
  } else {
    s[a[a$type == "entity"]]
  }
}

entities(text_doc, kind = "person")

I found an easier way instead of annotation.
You could get rid of the punctuations then use:

  text = unlist (strsplit (text, split = " "))

Or what do you think?

unlist is easier yes, can be used where one doesn't really need punctuations in their analysis.. But would not advice on using it if you're looking to doing some sentiment analysis since punctuations like exclamation marks(!) etc. are needed.

Hi CateGitau,
I tried NER.r but got "error in annotations(text): can't find function "annotations"
Traceback:

1. entities(text_doc, kind = "person")

the complete output was:
[1] "My name is Catherine Gitau, I work at Ongair Limited in Nairobi, Kenya"

id type start end features
1 sentence 1 70 constituents=<<integer,15>>
2 word 1 2
3 word 4 7
4 word 9 10
5 word 12 20
6 word 22 26

'My' 'name' 'is' 'Catherine' 'Gitau' ',' 'I' 'work' 'at' 'Ongair' 'Limited' 'in' 'Nairobi' ',' 'Kenya' 

Error in annotations(text): non trovo la funzione "annotations"
Traceback:

1. entities(text_doc, kind = "person")

Change annotations to annotation.

a <- annotations(text)[[1]]
a <- annotation(text)[[1]]

Ref: https://www.rdocumentation.org/packages/NLP/versions/0.2-1/topics/AnnotatedPlainTextDocument

Hi Catherine, thank you for the code you presented above. It helps a lot. However, when I run the code, I obtained the following results

entities(text_doc, kind = "person")
character(0)
entities(text_doc, kind = "location")
character(0)
entities(text_doc, kind = "organization")
character(0)

In the annotations, see the following example results, I hope it worked well

text_doc <- AnnotatedPlainTextDocument(text, text_annotations)
words(text_doc)
[1] "c("" "Together"
[3] "we" "can"
[5] "stop" "Floods"
[7] "in" "#"
[9] "Dar_es_salaam" "https"
[11] "://t.co/MRTeSXYghq" "."
[13] "Let" "Stop"
[15] "Floods" "from"
[17] "our\ncommunities" "in"
[19] "#" "Dar_es_salaam"
[21] "https" "://t.co/LghNMt1C1v\n\nEarly"
[23] "Today" "."
[25] "Community" "engagement"
[27] "in" "Project"

Where do you think is the problem for the final result to be as character(0)?
I am working on a twitter data and would like to get place names and nouns mentioned in the tweets so that I can geocode the places.
Thank you for helpiing

For those looking the entities function does not work because annotations was deprecated, found the solution on a YouTube comment so thought I'd cross post for those searching. You will need to install and load the coreNLP library, update annotations to read annotation and remove the brackets after the document, this should then recognize and output the entities identified when running the function with a kind specified.
Install.packages("coreNLP")
library(coreNLP)
entities <- function(doc, kind) {
s <- doc$content
a <- annotation(doc)
if(hasArg(kind)) {
k <- sapply(a$features, [[, "kind")
s[a[k == kind]]
} else {
s[a[a$type == "entity"]]
}
}